@BraydenWilmoth Genuine question: the OpenAPI-generated TF provider is still uneven, and agents are more sensitive to that than humans. What's different in cf's pipeline? Also, GraphQL analytics isn't in OpenAPI at all, so I'd probably keep my in-house CLI for that.
Opus was absolute garbage today: failed two medium-complexity tasks, then dropped a Codex-made commit from master. When asked to double-check, it looked at logs showing the loss and still said everything was fine.
I feel like a model doing gradient descent: every day is different. Some days it’s 10 equally urgent background tasks and nonstop context switching — patching, approving, barely reading. Some days it’s deep focus plus codex-spark when the model lags.